CN1542695A

CN1542695A - Method, device and computer program for object representation and search on pattern

Info

Publication number: CN1542695A
Application number: CNA2004100322393A
Authority: CN
Inventors: Mz; M·Z·波伯
Original assignee: Mitsubishi Electric Corp
Current assignee: Rakuten Group Inc
Priority date: 1999-07-05
Filing date: 2000-07-03
Publication date: 2004-11-03
Anticipated expiration: 2020-07-03
Also published as: GB2351826A; US7356203B2; GB2393012A; CN1295649C; US20080065610A1; BR0006894A; GB2391678B; GB2391677A; JP2011100467A; KR20030078952A; GB0325153D0; US7483594B2; US7162105B2; GB2391677B; KR100708799B1; CN1967542A; US20070065016A1; KR100431677B1; GB2391676A; JP4875200B2

Abstract

A method of representing an object appearing in a still or video image, by processing signals corresponding to the image, comprises deriving a plurality of numerical values associated with features appearing on the outline of an object starting from an arbitrary point on the outline and applying a predetermined ordering to said values to arrive at a representation of the outline.

Description

Method for displaying or searching target in image, device and computer system thereof

Technical Field

The present invention relates to the display of objects appearing in still images or video images such as images stored in multimedia databases for the purpose of retrieval, and more particularly to a method and apparatus for searching for objects using such a display.

Background

In an application program of an image in an image library, an outline and a shape of an object appearing in a video image or a still image or a part of the object is displayed and saved efficiently. In a well-known method for indexing and retrieving the additional shape library, a Curvature Scale Space (CSS) display may be used. Details about CSS can be found in the papers "append effective shape indices using the robustness of the curvature scale space" (British Provisions of machine images pp.53-62, Edinburgh, British, 1996) and "append indices of image databases using the shape content of the focal scale space" (IEE Provisions of experts meeting on Intelligent databases, London, 1996). Both papers are written by Mokhtarian, s.abbasi and j.kittler, the contents of which are incorporated by reference in this specification.

In the CSS display, in order to obtain an outline of an object, a curvature function is used to display the outline from an arbitrary point on the outline. By performing a series of deformations for smoothing the shape, the shape of the contour is developed and the curvature function is studied. Further, specifically, zero crossings of the derivative function of the curvature function convolved with the family of gaussian filters are calculated. As a curvature scale space, zero crossings are plotted on the graph, as is well known. However, the x-axis is the normalized arc length of the curve and the y-axis is the unfolding parameter, particularly the parameter to which the filter is applied. The points on the graph form a ring of features representing the profile. The convex or concave portions constituting the contour of the object correspond to the ring of the CSS image. In CSS images, the ordinate of the peak of the most prominent ring is used for the display of the contour.

To search for an object holding an image in a database that matches the shape of the input object, CSS display of the input shape is calculated. The similarity between the input shape and the saved shape is judged by comparing the position and height of the peak of each CSS image using the matching algorithm language.

As a problem with the well-known CSS display, there is a problem that the peak of a specified profile is based on a curvature function calculated from an arbitrary point on the profile. When changing the starting point, it happens that the spikes of the CSS image are shifted periodically along the x-axis. Therefore, in calculating the similarity measure, all possible shifts have to be considered or at least the shift that occurs most easily has to be considered. As a result, the complexity of the search procedure and the matching procedure will increase.

It is therefore an object of the present invention to provide a method of representing an object appearing in a still or video image by processing a signal corresponding to the image, the method comprising the steps of deriving a plurality of numerical values relating to features appearing on the outline of the object starting from an arbitrary point on the outline and applying a specified classification to the values to obtain a display of the outline. Preferably, the value is derived from a CSS display of the profile, and the value preferably corresponds to a CSS peak.

As a result of the present invention, the calculation of the matching program can be greatly reduced without significantly reducing the search accuracy.

Disclosure of the invention

The method of displaying an object in an image according to claim 1 of the present invention is a method of representing an object appearing in an image by processing a signal corresponding to a still image or a video image, characterized in that: comprising the steps of deriving a plurality of numerical values relating to features appearing on the contour starting from an arbitrary point on the contour of the object and applying a specified classification to the values to obtain a display of the contour.

The method of displaying an object in an image according to claim 2 of the present invention is characterized in that: the resulting display is classified as specified regardless of the starting point on the outline.

The method of displaying an object in an image according to claim 3 of the present invention is characterized in that: the values reflect the bending points on the curve.

The method of displaying an object in an image according to claim 4 of the present invention is characterized in that: the display of the curvature scale space of the contour is obtained by generating a plurality of contour curves by smoothing the contour in a plurality of stages using the smoothing parameter σ, deriving a curve representing the characteristics of the original contour using values representing the maximum value and the minimum value of the curvature of each contour curve, and selecting the ordinate of the peak of the curve representing the characteristics as numerical values.

The method of displaying an object in an image according to claim 5 of the present invention is characterized in that: the ordinate of the curve representing the feature corresponds to the arc length parameter and the smoothing parameter of the profile.

The method of displaying an object in an image according to claim 6 of the present invention is characterized in that: and classifying the ordinate value of the peak according to the value of the height of the peak corresponding to the smoothing parameter.

The method of displaying an object in an image according to claim 7 of the present invention is characterized in that: values are sorted starting from the maximum.

The method of displaying an object in an image according to claim 8 of the present invention is characterized in that: the values are sorted in order of decreasing size.

The method of displaying an object in an image according to claim 9 of the present invention is characterized in that: values are sorted starting from the minimum.

The method of displaying an object in an image according to claim 10 of the present invention is a method of representing an object appearing in an image by processing a signal corresponding to a still image or a video image, characterized in that: comprising a step of deriving a plurality of numerical values relating to the features appearing on the contour in order to represent the contour of the object and a step of deriving a coefficient representing the reliability of the display using a relationship between at least 2 of the numerical values.

The method of displaying an object in an image according to claim 11 of the present invention is characterized in that: the coefficients are based on the ratio between 2 of the values.

The method of displaying an object in an image according to claim 12 of the present invention is characterized in that: the above ratio is the ratio of 2 maxima.

The method of displaying an object in an image according to claim 13 of the present invention is characterized in that: the display of the curvature scale space of the contour is obtained by generating a plurality of contour curves by smoothing the contour in a plurality of stages using the smoothing parameter σ, by using values representing the maximum value and the minimum value of the curvature of each contour curve for deriving a curve representing the feature of the original contour, and by selecting the ordinate of the peak of the curve representing the feature as a numerical value.

The method of displaying an object in an image according to claim 14 of the present invention is characterized in that: deriving the value using the method of any one of claims 1 to 9.

The method of retrieving an object in an image according to claim 15 of the present invention is a method of retrieving an object in an image by processing a signal corresponding to a still image or a video image, characterized in that: comprising the steps of inputting a query in the form of a 2-dimensional outline, deriving descriptors of the outline using the method of any one of claims 1 to 9, taking descriptors of objects in the saved image derived using the method of any one of claims 1 to 9 and comparing each descriptor of the saved objects with the query descriptor, and selecting and displaying at least 1 result corresponding to an image containing an object representing the degree of similarity between the query and the object based on the comparison.

The method of retrieving an object in an image according to claim 16 of the present invention is characterized in that: deriving coefficients for the queried contour and each saved contour using the method of any of claims 10-12, comparing using only the specified classification or using the specified classification and some other classification related to the coefficients.

The method of displaying an object in an image according to claim 17 of the present invention is a method of representing a plurality of objects appearing in an image by processing a signal corresponding to a still image or a video image, characterized in that: the method includes a step of deriving a plurality of numerical values associated with features appearing on the contours of the respective objects and a step of obtaining a display of the respective contours by applying a specified classification identical to a value representing the respective contours.

The apparatus for displaying or retrieving an object in an image according to claim 18 of the present invention is characterized in that: adapted to perform the method of any of claims 1 to 17.

The computer program of claim 19 of the present invention for displaying or retrieving an object in an image is characterized in that: adapted to perform the method of any of claims 1 to 17.

The computer system for displaying or retrieving an object in an image according to claim 20 of the present invention is characterized in that: programmed to act according to the method of any of claims 1 to 17.

The computer-readable storage medium of claim 21 of the present invention, wherein: storing computer executable processes for implementing the method of any of claims 1 to 17.

The method of displaying an object in an image according to claim 22 of the present invention is characterized in that: an object in a still image or a video image is displayed substantially the same as that described in the present specification with reference to the drawings.

The method of retrieving an object in an image according to claim 23 of the present invention is characterized in that: as is substantially the same as described in this specification with reference to the drawings, an object in a still image or a video image is retrieved.

The computer system for displaying or retrieving an object in an image according to claim 24 of the present invention, wherein: substantially as herein described with reference to the accompanying drawings.

Brief description of the drawings

FIG. 1 is a block diagram of a video database system.

Fig. 2 is a diagram of the contour of an object.

Fig. 3 is a diagram of a CSS display showing the profile of fig. 2.

Fig. 4 is a block diagram showing a search method.

Best mode for carrying out the invention

Hereinafter, embodiments of the present invention will be described with reference to the drawings.

Example 1.

FIG. 1 shows a video database system for computer processing according to an embodiment of the present invention. In this system, a control device 2 in the form of a computer, a display device 4 in the form of a monitor, a pointing device 6 in the form of a mouse, an image database 8 containing stored still images and video images, and a descriptor database 10 storing descriptors of objects or parts of objects appearing in images stored in the image database 8 are included.

Descriptors representing the shapes of objects of interest appearing in the images of the image database are derived by the control device 2 and stored in the descriptor database 10. The control device 2 operates under control of an appropriate program for executing the method described below, and derives a descriptor.

1, for a contour of a specified object, a CSS display of the contour is derived.

The CSS display is performed using a well-known method described in one of the above papers.

Specifically, the contour is expressed by mapping expression Ψ { (x (u), y (u), u ∈ [0, 1]) } (where u is a normalized arc length parameter).

The profile is smoothed by convolution (convolute) with an ID gaussian kernel g (u, p), and for variations in p the curvature zero crossings of the developed (evolving) curve are examined. The zero crossings are specified using the following formula representing the curvature. Namely, it is

Wherein,

X(u，σ)＝x(u)*g(u，σ) Y(u，σ)＝y(u)*g(u，σ)

and,

X_u(u，σ)＝x(u)*g_u(u，σ) X_uu(u，σ)＝x(u)*g_uu(u，σ)

in the above formula, a represents convolution and a subscript represents a derivative function.

The number of curvature zero crossings varies with ρ, and when ρ is very high, Ψ becomes a convex curve of the zero crossings.

The zero crossing point (u, ρ) is depicted on the graph as CSS image space. As a result, a curve representing the characteristics of the original contour is formed. Then, a peak of a curve representing the feature is specified, and a corresponding ordinate is extracted and stored. Typically, the above results give a set of n coordinate pairs ((x1, y1), (x2, y2), … (xn, yn)) where n is the number of spikes, xi is the location of the arc length of the ith spike, yi is the height of the spike.

The classification and position of the curve representing the feature and the corresponding spike appear in the CSS image space in relation to the starting point of the curvature function described above. In the present invention, the ordinate of the peak is reclassified using a dedicated classification function.

The 1 mapping T is sorted by the spike index {1 … n } and the 1 of the index {1 … n } of the new set.

In the present embodiment, the pair of ordinates is classified by considering the size of the y ordinate. The largest spike is selected 1. Assume that the kth peak is most prominent. In this case, (xk, yk) is the 1 st coordinate in the numerically classified set. In other words, t (k) is 1. Similarly, the ordinate of the other peaks is reclassified in order of decreasing peak height. When the 2 peaks have the same height, the peak having the x coordinate closest to the x coordinate of the above-described ordinate pair is arranged as the 1 st peak. In other words, each ordinate pair having the original index i is assigned to a new index. However, t (i) ═ j, and yj > ═ y (j + 1). In addition, each value xi corresponds to a shift of the period of-xk.

As a specific example, the result of the CSS image shown in FIG. 3 can be obtained from the contour shown in FIG. 2. Details of the ordinate of the peak of the curve of the CSS image are shown in table 1 below.

TABLE 1

Spike index	X	Y
Spike index	X	Y	1	0.124	123
2	0.68	548	1	0.124	123
2	0.68	548	3	0.22	2120
4	0.773	1001	3	0.22	2120
4	0.773	1001	5	0.901	678

These spikes are classified using the classification method described above. That is, the ordinate classifies in order of decreasing height of the spikes. In addition, the x-ordinate bites in the direction of zero by an amount equal to the original x-ordinate of the maximum peak. As a result, the re-classified peak coordinates shown in table 2 below were formed.

TABLE 2

Spike index	X	Y
Spike index	X	Y	1	0	2120
2	0.553	1001	1	0	2120
2	0.553	1001	3	0.681	678
4	0.46	548	3	0.681	678
4	0.46	548	5	0.904	123

Using the vertical coordinates of these re-classified peaks, a database of descriptors is formed of the contours of the objects stored in the data 10. In the present embodiment, the ordinate of the peak is stored in the sort order shown in table 2. Alternatively, the ordinate may be stored together with an associated index indicating a new sort order.

Example 2.

Next, an alternative method of expressing the contour of the object of example 2 is explained.

The CSS display representing the profile is derived as described above. However, the classification of the ordinate of the spike is different from that of embodiment 1 described above. Further, specifically, 1 st, the maximum spike is selected. Assume that peak k is the most prominent peak. At this time, (xk, yk) becomes the 1 st peak in the sorted set of peaks. The ordinate of the peak of the subsequent peak with respect to the original index i is t (i) ═ J and xj < ═ x (J + 1). All values xi are bitten downward by an amount xk equal to the original x-ordinate of the original peak k.

In other words, in the method of embodiment 2, the largest peak is selected and allocated to the 1 st bit, and then the remaining peaks are allocated in the original order from the largest peak.

Table 3 below shows a table of peaks in table 1 classified according to example 2.

TABLE 3

Spike index	X	Y
Spike index	X	Y	1	0	2120
2	0.46	548	1	0	2120
2	0.46	548	3	0.553	1001
4	0.681	678	3	0.553	1001
4	0.681	678	5	0.904	123

In the development of the above-described embodiments 1 and 2, the reliability Coefficient (CF) is associated with each display of the shape. CF is calculated from the ratio of the 2 nd maximum peak to the maximum peak of the specified shape.

For the profile shown in fig. 2, the CF value is CF 1001/2120. In this example, the quantization process is performed by bringing the CF closest to 0.1, thereby reducing the storage requirement. Therefore, in this example, CF is 0.5.

The CF value for this example is a reflection of the degree of scab displayed, i.e., uniqueness. In this example, a CF value close to 1 means that the reliability is low, and a CF value close to zero indicates that the reliability is high. In other words, the closer the 2 maximum peaks are, the less likely it is that the display is correct.

When the matching procedure described below is performed, the CF value can be a useful value.

Example 3.

A method of retrieving an object in an image according to an embodiment of the present invention is described below with reference to fig. 4, which is a block diagram representing a retrieval method.

In this example, the descriptor derived according to the above-described classification method 1 is stored in the descriptor database 10 of the system of fig. 1 together with the associated CF value.

The user starts the search by tracing the outline of the object on the display using the pointing device (step 410). Next, the control device 2 derives a CSS display of the input contour, and classifies the vertical coordinates of the peaks according to the same classification function as that used for the image in the database, to obtain a descriptor indicating the input contour (step 420). Then, the control device 2 calculates the CF value of the input profile by calculating the ratio of the 2 nd maximum peak to the maximum peak, and performs quantization processing of the result (step 430).

Then, the control device 2 compares the C F value of the input contour with a specified threshold value (step 440). In this example, the threshold is 0.75. Representing a relatively high reliability on the accuracy of the input descriptors, the next step is the step of considering the CF values of the model under consideration (i.e. the image saved in the database) when the CF values are lower than the threshold. When the CF of the model is still below the threshold (step 450), a comparison of the input descriptors to the model is made using only the descriptors in the specified sort order (step 460). When the CF of the input descriptor or model is greater than the threshold, a match is made by comparing all possible different sort orders of ordinate values in the input descriptor with the model descriptors in the database (step 470).

In the database, the similarity measurement values of the descriptors are compared by matching using an appropriate algorithm language obtained as a result. A language of well-known matching algorithms as described in the above-mentioned paper may also be used. The matching sequence will be briefly described below.

Given the shape of the 2 closed contours, the image curve Ψ i and the model curve Ψ m and the set values of the peaks of these curves { (xi1, yi1), (xi2, yi2),., (xin, yin) } and { (xm1, ym1), (xm2, ym2),., (xmn, ymn) }, a similarity measure can be calculated. The similarity measure is defined as the total cost of matching the spike in the image with the spike in the model. The match that minimizes the total cost is calculated using dynamic programming. And feeding the peak obtained from the model back to the peak obtained from the image for matching by using the algorithm language, and calculating various matched costs. The peaks of each model may be matched to unique image peaks, or the peaks of each image may be matched to unique model peaks. There may be peaks in the model and/or image that are still unmatched, and there is an additional penalty for unmatched peaks. When the horizontal distance of the 2 spikes is less than 0.2, the 2 spikes can be matched. The cost of matching is the linear length of the 2 matched spikes. The cost of an unmatched spike is its height.

More specifically, the algorithmic language works by creating and expanding a tree-like structure corresponding to a peak of a node match.

1. A start node is created which is composed of the maximum value of the image (xik, yik) and the maximum value of the model (xir, yir).

2. Additional start nodes are created for each remaining model peak within 80% of the maximum value of the image peak.

3. The cost of each start node created in 1 and 2 above is initialized to the absolute value of the difference between the start node and the y-coordinates of the linked image spike and model spike.

4. The CSS shift parameter α defined as the difference between the x (horizontal) coordinates of the model peak and the image peak matched at the start node is calculated for each of the start nodes in the above 3. The shift parameters are different for each node.

5. A list of model peaks and a list of image peaks are made for each starting node. Information about which peaks have not yet matched is contained in the list. For each starting node, for "matched", the peak matched at that node is flagged, and all other peaks are flagged as "unmatched".

6. The lowest cost node is fed back until the following condition 8 is satisfied (the child nodes of each node are located after the nodes created in steps 1 to 6). To enlarge the node, the following steps are used.

7. Expansion of nodes

The maximum value of the largest unmatched scale image curve CSS is selected (xip, yip) when there are at least 1 more than 1 model spike of the still unmatched images. The maximum value selected by applying the start node shift parameter (calculated in step 4) is mapped onto the model CSS image, the selected spike having the coordinates (xip-alpha, yip). The closest model curve peak (xms, yms) that did not match was determined. When the horizontal distance between 2 peaks is less than 0.2 (i.e. | xip-alpha-xms | < 0.2), 2 peaks are matched, and the matching cost is defined as the length of the straight line between 2 peaks. The matched cost is added to the total cost of the node. For matching spikes, the matched spike is removed from each list by adding a flag as "matched". At horizontal distances between 2 peaks greater than 0.2, the image peaks (xip, yip) are not matched. At this point, the height yip of the image spike is added to the total cost, and the spike (xip, yip) is removed from the image spike table by flagging the "matched" spike.

When the above condition is not met (only unmatched image spikes or only unmatched model spikes are present), the model is placed in a still unmatched state.

As the heightened height of the unmatched image spike or model spike, the cost of the match is defined and the spike is removed from the list.

8. After the 7-point expansion, when there is no unmatched peak in both the image list and the model list, the matching process is terminated. The cost of this node is a similarity measure between the image and the model curve. In the presence of spikes, return to 7 above, enlarging the lowest cost node.

And exchanging the peak value of the image curve with the peak value of the model curve, and repeating the steps. The final matching value is the lower of the 2 peaks.

As another example, for each position in the sorted order, the distance between the input x value and the x value of the model corresponding thereto and the distance between the input y value and the y value of the model corresponding thereto are calculated. And calculating the sum distance of all the positions, wherein the smaller the sum distance is, the closer the matching degree is. When the input profile is different from the tree of peaks of the model, the heights of the remaining unmatched peaks are contained in the resultant distance.

The above steps are repeated for each model of the database (step 480).

The similarity values resulting from the results of the match comparison are sorted (step 490) and the target corresponding to the medium character having the similarity value representing the most recent match value (i.e., the lowest similarity value in this example) is then displayed to the user on the display device 4 (step 500). The target tree of display objects may be preset or selected by the user.

In the above embodiment, when the CF value is greater than the threshold, all possible orders of the input descriptor values at the time of matching are considered. However, it is also possible to not have to consider all possible orders, but instead only a few possible orders of the shifts of all periods of a few monkeys as shown in the original CSS. Further, in the above-described embodiment, the threshold value is set to 0.75, but the threshold value may be set to a different level. For example, when the threshold is set to zero, all matches are made by analysis of several or all possible sequences. Therefore, although the amount of calculation required is increased as compared with the case where the threshold value is larger than zero, the peak is classified, and the x-coordinate of the peak is adjusted to a specific start point or target rotation, so that the amount of calculation required is reduced as compared with the original system in which such adjustment is not performed. Therefore, by setting the threshold to zero, it is possible to reduce a number of calculations, and the retrieval performance is exactly the same as that of the original system.

Alternatively, when the threshold is set to 1, matching is performed using only the stored order. In this case, the search accuracy is slightly lowered, and the required amount of calculation can be significantly reduced.

Various modifications can be made to the above embodiment. For example, instead of the classification of the total coordinate values of the CSS peaks described in embodiment 1 and embodiment 2, another classification method may be used. For example, the total coordinate values may be arranged in the order of increasing the height of the peak, instead of being arranged in the order of decreasing the height of the peak. Instead of storing the sorted values in a database, the sorting may also be performed in the matching step.

Possibility of industrial utilization

The system of the present invention may be provided in, for example, an image library. Alternatively, the database may be connected to the control device via a network such as the internet by a temporary link of a telephone line, and may be disposed at a remote place from the control device of the system. For example, the image database and the descriptor database may be provided in a permanent storage device or a portable storage medium such as a ROM or a DVD.

The components of the system described above may be provided in the form of software or hardware. The invention has been described above in the form of a computer system, but the invention can also be implemented in other forms using dedicated chips or the like.

While specific examples of the method of representing the 2D shape of the object and the method of calculating the value representing the similarity between the 2 shapes have been given above, any appropriate method may be used as well.

For example, matching of the target image or performing loop filtering for verification purposes may also use the present invention.

Claims

1. A method of retrieving an object in an image or a series of images by processing signals corresponding to said image or series of images, the method comprising: the method comprises the steps of inputting a query in the form of a 2-dimensional profile, deriving descriptors of the profile, acquiring descriptors of objects in a stored image, comparing the query descriptors with the descriptors of the stored objects, and selecting and displaying at least 1 result corresponding to an image containing the objects whose comparison indicates the degree of similarity between the query and the objects, wherein the stored and query descriptions are derived from a curvature scale space display of the profile of the objects, and the peak ordinate values displayed by the CSS are classified according to the peak height values of the peak ordinate.

2. The method of claim 1, wherein: the above values are sorted starting from the maximum value.

3. The method of claim 2, wherein: the above values are sorted in order of decreasing size.

4. The method of claim 1, wherein: the above values are sorted starting from the minimum value.

5. The method of any of claims 1 to 4, wherein: deriving coefficients for the queried contour and each saved contour using only a specified classification or using the specified classification and some other classification related to the coefficients, wherein the coefficients are obtained by deriving a plurality of numerical values for displaying the contour associated with a feature appearing on the contour of the object and deriving a coefficient indicating the reliability of the display using a relationship of at least two of the numerical values.

6. The method of claim 5, wherein: the coefficients are based on ratios between 2 of the values.

7. The method of claim 6, wherein: the above ratio is the ratio of 2 maxima.

8. A method according to any one of claims 5 to 7, wherein the numerical value corresponds to a peak ordinate value in a CSS display.

9. The method of claim 8, wherein said peak ordinate values are each classified according to a peak height value.

10. A method for retrieving an object appearing in an image, comprising:

receiving an input comprising at least one contoured object;

determining a curvature scale space display of the input profile to produce a peak ordinate of the curvature scale space display;

classifying the spike ordinate according to a spike height value to produce a shape descriptor of the input profile;

comparing the shape descriptor of the input outline with shape descriptors of images stored in memory to find at least one substantially matching image; and is

Outputting said at least one substantially matched image.

11. A system for retrieving objects appearing in an image, comprising:

input means for inputting at least one contoured object;

a controller and detector for receiving said contoured input target and determining a curvature scale space representation of said input contour to produce a peak ordinate of the curvature scale space representation;

wherein the controller classifies the spike ordinate according to a spike height value to generate a shape descriptor of the input profile;

a comparator for comparing the shape descriptor of the input outline with shape descriptors of images stored in a memory to find at least one substantially matching image; and

a display for displaying the at least one substantially matching image.

12. A method for retrieving an image for use in a comparator in the retrieval system of claim 11, comprising:

receiving a first shape descriptor of a target in the image, the shape descriptor including a plurality of spike ordinates of a Curvature Scale Space (CSS) display of the target, the spike ordinates being classified according to a spike height value, and the spike ordinates of the CSS display being generated by smoothing the contour in a plurality of stages; and is

Comparing the first shape descriptor of the target with second shape descriptors of images stored in memory, finding at least one substantially matching image, wherein for each of the spike ordinates classified according to spike height, a comparison is made with corresponding coordinates of the second shape descriptors in the memory.

13. A method for retrieving an image, of a controller used in the retrieval system of claim 11, comprising:

generating a spike ordinate of a Curvature Scale Space (CSS) display of a target profile by smoothing the CSS display in multiple stages;

classifying the spike ordinate according to a spike height value to produce a shape descriptor of the target;

passing the shape descriptor onto a comparator to retrieve at least one substantially matching image; and is

Receiving said one substantially matching image from said comparator and displaying it.

14. An apparatus adapted to carry out the method of any one of claims 1 to 10.